Miniaturization

The Analog Devices jfet-based references, ADR4xx are great. But not cheap.

Yes ok that is cheaper than I am used to, but there is a substantial premium for the 0.1% tolerance, let alone 0.01%! Also nice that they are thin-film so hopefully not too much excess noise.

Yes, much like on a chip, though the prices I have seen are a dollar or two. On a chip you do sometimes need to worry about thermal gradients affecting resistors as the tempco can be high for polysilicon, so interleaving them is sometimes worthwhile and easy to do.

That is true.

Hmm the LM329 is 75nV/rt-Hz at 6.9V, so I guess -159.3dB/Hz relative to the DC "signal", using 0.5mA*6.9V = 3.45mW. I did a bandgap-based LDO that had about 20nV/rt-Hz on its 2.45V output, so -161.7dB/Hz by the same metric, using about 1mA of idle current from a 2.7V rail, so 2.7mW. Pretty similar I suppose.

I have not found them. NP0 and film types are about the only ones that I would consider "super good" and they are big and usually expensive in the sizes that would make a good 1Hz filter. I would not call electrolytics or high-K ceramics "super good". BTW we came up with a trick to avoid the cost and size of off-chip big capacitors and still make a low-noise filter for the PLLs, see US6891412.

Yes, it is better to make the reference quiet so that it doesn't need filtering. Then start-up time is not a problem.

Sounds like it should help (except for acoustic waves at ultrasonic frequencies maybe) but also a hassle (and uses a lot of space). I have to admit that I have never personally experienced microphony from high-K caps, but then I don't use them in low-level signal paths.

Quite true. If one has the option to do parts of a system in whichever domain is more suitable, that should result in the best product.

Yes we did stuff like that with PLLs where you want it to lock fast and then have low noise. Often the tricky part is not kicking it too much when you switch modes.

"Jim Thompson" wrote in message news: snipped-for-privacy@4ax.com...

In what quantities? Can you get me qty 10 for that price? Maybe only qty

As many times as I have to repeat it, it never seems to sink in... I'm talking PROTOTYPE QUANTITIES in this thread.

If you can't bend your brain to contemplate creating something, economically, in qty < 10k, I don't need your opinion.

Tim

Yeah, people are always in need of TRNGs!

Tim

Tim, You are getting tiring. You have no clue of the real world of chips and are spouting nonsense. Most of my client quantities are :-}

(*) Although my LED billboard driver is at >1 Million/month. ...Jim Thompson

"Jim Thompson" wrote in message news: snipped-for-privacy@4ax.com...

So you've done turnkey(ish) designs for, say, $1250, all said and done?

That's unheard of, from anything anyone else in this thread has said, or anything I've seen elsewhere online... If it's true, you could've just said so.

Or if those two numbers /don't/ go together, then I suggest you stop spouting nonsense and retake your basic arithmetic class. >:-}

Tim

If I said that small chips on an ordinary cheap process can sell for $25, and someone sells them to a customer for that price and the customer pays it, then it pretty much does. It might not prove a bunch of other different things that I and several people are saying in this thread.

Good, but hopefully someone is buying them for more than they cost to make. If the function of the chip is much better than its competitors (or there are no competitors) then it could sell for quite a bit more. Of course that depends on the market segment as some products face tough competition and some don't.

I'll agree with that!

Right, I meant 0.1%. For instance:

ERA-3AEB103V Panasonic Electronic Components RES SMD 10K OHM 0.1% 1/10W 0603 315,000 - Immediate

The 0.01% ones are more like half a buck--not much point in making a

Yeah, common-centroid helps a lot.

That's very interesting. How did you get the noise that low? Was this a bipolar bandgap?

I do a lot of ultrasensitive instruments, and as long as I keep clear of the super tiny ones, e.g. 10 uF 0603 (or even 0402!), and don't use anything uglier than X7R, I've never had a serious problem with microphonics. Of course most of my stuff doesn't get banged around too much. Y5V and Z5U and their ilk are definitely bad news.

Looks interesting--I'll have a squint at it. A quick skim revealed that you let your lawyer write the specification, so it's not an easy read. ;)

At one point it claims that the noise is lower than that of an equivalent RC filter, but doesn't (as far as I've yet read) mention kT/C noise.

Yup.

Cheers

Phil Hobbs

I'll fart twice for $1250 >:-}

A typical design can be as low as $25K circuit design, $12K for layout, $20K for first round of processing (includes masks).

If your circuit design capability is limited to seven transistors...

(Actually, my first OpAmp design, ~1963-64, was only 16 transistors, and is still in production by Lansdale... MC1530-31) ...Jim Thompson

I use the same tactics in other fields. If they are so shy about their price or the quality of their own products, they most surely have reasons to do so -- and it is the clearest message possible to avoid them.

Best regards, Piotr

(*) PPS: 10K ANNUAL ...Jim Thompson

The clients they lose this way are the low volume, high effort customers that are the least profitable. We would like to think we are all worth supporting, but the truth is we aren't. I'm always happy when I can get the attention of a vendor.

History says that's wrong. Carnegie signed on to a very demanding contract, with Eads, on steel for a bridge across the Mississippi. The delivered steel had to pass testing to a specified Young's modulus, and lots of product got returned.

Carnegie didn't make money on that contract, but he DID improve his steel-making so that his products set the standard for decades afterward. It was a sound, indeed a brilliant, decision, even if it didn't enhance that quarter's balance sheet.

_Dams and Other Disasters_ by Arthur Morgan, is a good read on this subject.

I don't see how the example relates .

Right. We've done very small quantities of chips through MOSIS. We've ordered anywhere from 100 to 400 of any particular design. Depending on the size (very roughly 3 x 5 mm up to 5 x 8 mm) we've had costs in the neighborhood of $18,000 to $40,000 for these runs. (Now, when we actually have a home run and the chips not only work well and meet a wider need) we can go back to MOSIS and have them open up a spare wafer and get more chips for a song - maybe $8000 for a couple hundred more parts.

I've never heard of anybody getting chips made for a few thousand $ total cost, except the educational runs at MOSIS, which are heavily subsidized.

Jon

We use tons of the Susumu thinflms. 0.1% cost around a dime, and 0.05% is around 30 cents. I temperature tested a few and saw numbers like 4 to 6 PPM/degC, hard to measure.

We usually software calibrate things, so tempco is more important to us than initial accuracy.

Yes, bipolar. One important trick is not being afraid to burn most of a milliamp in the core, which few "low noise" reference designers have dared to do for some reason. The original bandgap that I adapted into an LDO was designed by my boss at the time, who really knew what he was doing, the second inventor on US4833344. That patent is not for the bandgap but if you look at figure 2 in it, and add a resistor from collector of Q2 to the top rail (top of R1), that is the guts of it as far as I can remember. Q2 is *much* bigger than Q1, and runs at several times lower current (due to choice of Q2's collector resistor). You need to add some feedback arrangement that adjusts the voltage at the top rail until Vce of Q2 is one Vbe, and there are simple, quiet ways to do that. With the right choice of resistor values the top rail should be at the bandgap voltage.

Just look at the pictures is my suggestion. When I can no longer understand the text myself, I know it is ready to file!

Yes the objective was to avoid the resistor noise, as that was the thing preventing the capacitors from being small enough to integrate. I have never had a good understanding of kT/C noise. To the best of my little knowledge, that noise does not appear across a capacitor that has one terminal floating, but does appear across a capacitor when it is used with an analog switch to form a sample-and-hold, e.g. in a switched capacitor filter or a charge redistribution ADC. It is my (poor) understanding that if you made one of these sample-and-holds take a sample of e.g. ground, the noise energy would originate from the Johnson noise of the analog switch as it transitions from roughly a short-circuit to roughly an open-circuit because the resistance goes through values where it is noisy as it switches. I'm not sure how that would affect the filter that we patented, I suspect it would depend on details of the charge pump. It certainly didn't mess things up to the extent of being a problem, though that may be due to the rather careful design of the charge pump by the other inventor.

By the way, at the cost of using active devices, I have heard that there tricky are ways around kT/C noise even in sample-and-hold circuits, e.g.:

Chris

Sure, correlated double sampling for instance.

The energy on a capacitor (CV**2)/2 is a single classical degree of freedom, so by equipartition it has a mean of kT/2. It doesn't matter how it gets there--a switch, a resistor, air ions....

That's actually the beginning of the simplest derivation of the Johnson noise formula that I know of.

Cheers

Phil Hobbs

That scheme is as old as the hills. I believe it's one of Widlar's ideas, though it could be Jim Williams or Bob Dobkin.